Replacement of the buffer replacement strategy with an ARC

algorithm adopted for PostgreSQL.

Jan

src/backend/commands/vacuum.c

@@ -13,7 +13,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/commands/vacuum.c,v 1.265 2003/11/12 21:15:51 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/commands/vacuum.c,v 1.266 2003/11/13 00:40:00 wieck Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -33,6 +33,7 @@
 #include "commands/vacuum.h"
 #include "executor/executor.h"
 #include "miscadmin.h"
+#include "storage/buf_internals.h"
 #include "storage/freespace.h"
 #include "storage/sinval.h"
 #include "storage/smgr.h"
@@ -310,8 +311,16 @@ vacuum(VacuumStmt *vacstmt)
 			else
 				old_context = MemoryContextSwitchTo(anl_context);
 
+			/*
+			 * Tell the buffer replacement strategy that vacuum is
+			 * causing the IO
+			 */
+			StrategyHintVacuum(true);
+
 			analyze_rel(relid, vacstmt);
 
+			StrategyHintVacuum(false);
+
 			if (vacstmt->vacuum)
 				CommitTransactionCommand();
 			else
@@ -749,6 +758,12 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, char expected_relkind)
 	SetQuerySnapshot();			/* might be needed for functions in
 								 * indexes */
 
+	/*
+	 * Tell the cache replacement strategy that vacuum is causing
+	 * all following IO
+	 */
+	StrategyHintVacuum(true);
+
 	/*
 	 * Check for user-requested abort.	Note we want this to be inside a
 	 * transaction, so xact.c doesn't issue useless WARNING.
@@ -763,6 +778,7 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, char expected_relkind)
 							  ObjectIdGetDatum(relid),
 							  0, 0, 0))
 	{
+		StrategyHintVacuum(false);
 		CommitTransactionCommand();
 		return true;			/* okay 'cause no data there */
 	}
@@ -796,6 +812,7 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, char expected_relkind)
 				(errmsg("skipping \"%s\" --- only table or database owner can vacuum it",
 						RelationGetRelationName(onerel))));
 		relation_close(onerel, lmode);
+		StrategyHintVacuum(false);
 		CommitTransactionCommand();
 		return false;
 	}
@@ -810,6 +827,7 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, char expected_relkind)
 				(errmsg("skipping \"%s\" --- cannot vacuum indexes, views, or special system tables",
 						RelationGetRelationName(onerel))));
 		relation_close(onerel, lmode);
+		StrategyHintVacuum(false);
 		CommitTransactionCommand();
 		return false;
 	}
@@ -824,6 +842,7 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, char expected_relkind)
 	if (isOtherTempNamespace(RelationGetNamespace(onerel)))
 	{
 		relation_close(onerel, lmode);
+		StrategyHintVacuum(false);
 		CommitTransactionCommand();
 		return true;			/* assume no long-lived data in temp
 								 * tables */
@@ -863,6 +882,7 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, char expected_relkind)
 	/*
 	 * Complete the transaction and free all temporary memory used.
 	 */
+	StrategyHintVacuum(false);
 	CommitTransactionCommand();
 
 	/*

src/backend/storage/buffer/buf_init.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/buf_init.c,v 1.54 2003/08/04 02:40:03 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/buf_init.c,v 1.55 2003/11/13 00:40:01 wieck Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -48,9 +48,6 @@ long	   *CurTraceBuf;
 int			ShowPinTrace = 0;
 
 int			Data_Descriptors;
-int			Free_List_Descriptor;
-int			Lookup_List_Descriptor;
-int			Num_Descriptors;
 
 BufferDesc *BufferDescriptors;
 Block	   *BufferBlockPointers;
@@ -133,9 +130,6 @@ InitBufferPool(void)
 	int			i;
 
 	Data_Descriptors = NBuffers;
-	Free_List_Descriptor = Data_Descriptors;
-	Lookup_List_Descriptor = Data_Descriptors + 1;
-	Num_Descriptors = Data_Descriptors + 1;
 
 	/*
 	 * It's probably not really necessary to grab the lock --- if there's
@@ -156,7 +150,7 @@ InitBufferPool(void)
 
 	BufferDescriptors = (BufferDesc *)
 		ShmemInitStruct("Buffer Descriptors",
-					  Num_Descriptors * sizeof(BufferDesc), &foundDescs);
+					  Data_Descriptors * sizeof(BufferDesc), &foundDescs);
 
 	BufferBlocks = (char *)
 		ShmemInitStruct("Buffer Blocks",
@@ -176,16 +170,14 @@ InitBufferPool(void)
 		block = BufferBlocks;
 
 		/*
-		 * link the buffers into a circular, doubly-linked list to
-		 * initialize free list, and initialize the buffer headers. Still
-		 * don't know anything about replacement strategy in this file.
+		 * link the buffers into a single linked list. This will become the
+		 * LiFo list of unused buffers returned by StragegyGetBuffer().
 		 */
 		for (i = 0; i < Data_Descriptors; block += BLCKSZ, buf++, i++)
 		{
 			Assert(ShmemIsValid((unsigned long) block));
 
-			buf->freeNext = i + 1;
-			buf->freePrev = i - 1;
+			buf->bufNext = i + 1;
 
 			CLEAR_BUFFERTAG(&(buf->tag));
 			buf->buf_id = i;
@@ -199,14 +191,12 @@ InitBufferPool(void)
 			buf->wait_backend_id = 0;
 		}
 
-		/* close the circular queue */
-		BufferDescriptors[0].freePrev = Data_Descriptors - 1;
-		BufferDescriptors[Data_Descriptors - 1].freeNext = 0;
+		/* Correct last entry */
+		BufferDescriptors[Data_Descriptors - 1].bufNext = -1;
 	}
 
 	/* Init other shared buffer-management stuff */
-	InitBufTable();
-	InitFreeList(!foundDescs);
+	StrategyInitialize(!foundDescs);
 
 	LWLockRelease(BufMgrLock);
 }

src/backend/storage/buffer/buf_table.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/buf_table.c,v 1.29 2003/08/04 02:40:03 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/buf_table.c,v 1.30 2003/11/13 00:40:01 wieck Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -38,7 +38,7 @@ static HTAB *SharedBufHash;
  * Initialize shmem hash table for mapping buffers
  */
 void
-InitBufTable(void)
+InitBufTable(int size)
 {
 	HASHCTL		info;
 
@@ -50,7 +50,7 @@ InitBufTable(void)
 	info.hash = tag_hash;
 
 	SharedBufHash = ShmemInitHash("Shared Buffer Lookup Table",
-								  NBuffers, NBuffers,
+								  size, size,
 								  &info,
 								  HASH_ELEM | HASH_FUNCTION);
 
@@ -58,79 +58,63 @@ InitBufTable(void)
 		elog(FATAL, "could not initialize shared buffer hash table");
 }
 
-BufferDesc *
+/*
+ * BufTableLookup
+ */
+int
 BufTableLookup(BufferTag *tagPtr)
 {
 	BufferLookupEnt *result;
 
 	if (tagPtr->blockNum == P_NEW)
-		return NULL;
+		return -1;
 
 	result = (BufferLookupEnt *)
 		hash_search(SharedBufHash, (void *) tagPtr, HASH_FIND, NULL);
 	if (!result)
-		return NULL;
+		return -1;
 
-	return &(BufferDescriptors[result->id]);
+	return result->id;
 }
 
 /*
  * BufTableDelete
  */
 bool
-BufTableDelete(BufferDesc *buf)
+BufTableInsert(BufferTag *tagPtr, Buffer buf_id)
 {
 	BufferLookupEnt *result;
-
-	/*
-	 * buffer not initialized or has been removed from table already.
-	 * BM_DELETED keeps us from removing buffer twice.
-	 */
-	if (buf->flags & BM_DELETED)
-		return TRUE;
-
-	buf->flags |= BM_DELETED;
+	bool		found;
 
 	result = (BufferLookupEnt *)
-		hash_search(SharedBufHash, (void *) &(buf->tag), HASH_REMOVE, NULL);
+		hash_search(SharedBufHash, (void *) tagPtr, HASH_ENTER, &found);
 
-	if (!result)				/* shouldn't happen */
-		elog(ERROR, "shared buffer hash table corrupted");
+	if (!result)
+		ereport(ERROR,
+				(errcode(ERRCODE_OUT_OF_MEMORY),
+				 errmsg("out of shared memory")));
 
-	/*
-	 * Clear the buffer's tag.  This doesn't matter for the hash table,
-	 * since the buffer is already removed from it, but it ensures that
-	 * sequential searches through the buffer table won't think the buffer
-	 * is still valid for its old page.
-	 */
-	buf->tag.rnode.relNode = InvalidOid;
-	buf->tag.rnode.tblNode = InvalidOid;
+	if (found)					/* found something else in the table? */
+		elog(ERROR, "shared buffer hash table corrupted");
 
+	result->id = buf_id;
 	return TRUE;
 }
 
+/*
+ * BufTableDelete
+ */
 bool
-BufTableInsert(BufferDesc *buf)
+BufTableDelete(BufferTag *tagPtr)
 {
 	BufferLookupEnt *result;
-	bool		found;
-
-	/* cannot insert it twice */
-	Assert(buf->flags & BM_DELETED);
-	buf->flags &= ~(BM_DELETED);
 
 	result = (BufferLookupEnt *)
-		hash_search(SharedBufHash, (void *) &(buf->tag), HASH_ENTER, &found);
-
-	if (!result)
-		ereport(ERROR,
-				(errcode(ERRCODE_OUT_OF_MEMORY),
-				 errmsg("out of shared memory")));
+		hash_search(SharedBufHash, (void *) tagPtr, HASH_REMOVE, NULL);
 
-	if (found)					/* found something else in the table? */
+	if (!result)				/* shouldn't happen */
 		elog(ERROR, "shared buffer hash table corrupted");
 
-	result->id = buf->buf_id;
 	return TRUE;
 }
 

src/backend/storage/buffer/bufmgr.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.141 2003/09/25 06:58:01 petere Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.142 2003/11/13 00:40:01 wieck Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -260,12 +260,8 @@ ReadBufferInternal(Relation reln, BlockNumber blockNum,
 	if (status == SM_FAIL)
 	{
 		/* IO Failed.  cleanup the data structures and go home */
+		StrategyInvalidateBuffer(bufHdr);
 
-		if (!BufTableDelete(bufHdr))
-		{
-			LWLockRelease(BufMgrLock);
-			elog(FATAL, "buffer table broken after I/O error");
-		}
 		/* remember that BufferAlloc() pinned the buffer */
 		UnpinBuffer(bufHdr);
 
@@ -318,7 +314,7 @@ BufferAlloc(Relation reln,
 	INIT_BUFFERTAG(&newTag, reln, blockNum);
 
 	/* see if the block is in the buffer pool already */
-	buf = BufTableLookup(&newTag);
+	buf = StrategyBufferLookup(&newTag, false);
 	if (buf != NULL)
 	{
 		/*
@@ -379,7 +375,7 @@ BufferAlloc(Relation reln,
 	inProgress = FALSE;
 	for (buf = (BufferDesc *) NULL; buf == (BufferDesc *) NULL;)
 	{
-		buf = GetFreeBuffer();
+		buf = StrategyGetBuffer();
 
 		/* GetFreeBuffer will abort if it can't find a free buffer */
 		Assert(buf);
@@ -492,7 +488,7 @@ BufferAlloc(Relation reln,
 			 * we haven't gotten around to insert the new tag into the
 			 * buffer table. So we need to check here.		-ay 3/95
 			 */
-			buf2 = BufTableLookup(&newTag);
+			buf2 = StrategyBufferLookup(&newTag, true);
 			if (buf2 != NULL)
 			{
 				/*
@@ -535,29 +531,12 @@ BufferAlloc(Relation reln,
 	 */
 
 	/*
-	 * Change the name of the buffer in the lookup table:
-	 *
-	 * Need to update the lookup table before the read starts. If someone
-	 * comes along looking for the buffer while we are reading it in, we
-	 * don't want them to allocate a new buffer.  For the same reason, we
-	 * didn't want to erase the buf table entry for the buffer we were
-	 * writing back until now, either.
+	 * Tell the buffer replacement strategy that we are replacing the
+	 * buffer content. Then rename the buffer.
 	 */
-
-	if (!BufTableDelete(buf))
-	{
-		LWLockRelease(BufMgrLock);
-		elog(FATAL, "buffer wasn't in the buffer hash table");
-	}
-
+	StrategyReplaceBuffer(buf, reln, blockNum);
 	INIT_BUFFERTAG(&(buf->tag), reln, blockNum);
 
-	if (!BufTableInsert(buf))
-	{
-		LWLockRelease(BufMgrLock);
-		elog(FATAL, "buffer in buffer hash table twice");
-	}
-
 	/*
 	 * Buffer contents are currently invalid.  Have to mark IO IN PROGRESS
 	 * so no one fiddles with them until the read completes.  If this
@@ -709,13 +688,28 @@ BufferSync(void)
 	BufferDesc *bufHdr;
 	ErrorContextCallback errcontext;
 
+	int			num_buffer_dirty;
+	int		   *buffer_dirty;
+
 	/* Setup error traceback support for ereport() */
 	errcontext.callback = buffer_write_error_callback;
 	errcontext.arg = NULL;
 	errcontext.previous = error_context_stack;
 	error_context_stack = &errcontext;
 
-	for (i = 0, bufHdr = BufferDescriptors; i < NBuffers; i++, bufHdr++)
+	/*
+	 * Get a list of all currently dirty buffers and how many there are.
+	 * We do not flush buffers that get dirtied after we started. They
+	 * have to wait until the next checkpoint.
+	 */
+	buffer_dirty = (int *)palloc(NBuffers * sizeof(int));
+	num_buffer_dirty = 0;
+
+	LWLockAcquire(BufMgrLock, LW_EXCLUSIVE);
+	num_buffer_dirty = StrategyDirtyBufferList(buffer_dirty, NBuffers);
+	LWLockRelease(BufMgrLock);
+
+	for (i = 0; i < num_buffer_dirty; i++)
 	{
 		Buffer		buffer;
 		int			status;
@@ -723,10 +717,11 @@ BufferSync(void)
 		XLogRecPtr	recptr;
 		Relation	reln;
 
-		errcontext.arg = bufHdr;
-
 		LWLockAcquire(BufMgrLock, LW_EXCLUSIVE);
 
+		bufHdr = &BufferDescriptors[buffer_dirty[i]];
+		errcontext.arg = bufHdr;
+
 		if (!(bufHdr->flags & BM_VALID))
 		{
 			LWLockRelease(BufMgrLock);
@@ -855,6 +850,8 @@ BufferSync(void)
 			RelationDecrementReferenceCount(reln);
 	}
 
+	pfree(buffer_dirty);
+
 	/* Pop the error context stack */
 	error_context_stack = errcontext.previous;
 }
@@ -959,9 +956,9 @@ AtEOXact_Buffers(bool isCommit)
 
 			if (isCommit)
 				elog(WARNING,
-				"buffer refcount leak: [%03d] (freeNext=%d, freePrev=%d, "
+				"buffer refcount leak: [%03d] (bufNext=%d, "
 				  "rel=%u/%u, blockNum=%u, flags=0x%x, refcount=%d %ld)",
-					 i, buf->freeNext, buf->freePrev,
+					 i, buf->bufNext,
 					 buf->tag.rnode.tblNode, buf->tag.rnode.relNode,
 					 buf->tag.blockNum, buf->flags,
 					 buf->refcount, PrivateRefCount[i]);
@@ -1229,7 +1226,7 @@ recheck:
 			/*
 			 * And mark the buffer as no longer occupied by this rel.
 			 */
-			BufTableDelete(bufHdr);
+			StrategyInvalidateBuffer(bufHdr);
 		}
 	}
 
@@ -1295,7 +1292,7 @@ recheck:
 			/*
 			 * And mark the buffer as no longer occupied by this page.
 			 */
-			BufTableDelete(bufHdr);
+			StrategyInvalidateBuffer(bufHdr);
 		}
 	}
 
@@ -1543,7 +1540,7 @@ FlushRelationBuffers(Relation rel, BlockNumber firstDelBlock)
 				return -2;
 			}
 			if (bufHdr->tag.blockNum >= firstDelBlock)
-				BufTableDelete(bufHdr);
+				StrategyInvalidateBuffer(bufHdr);
 		}
 	}
 

src/backend/utils/misc/guc.c

@@ -10,7 +10,7 @@
  * Written by Peter Eisentraut <peter_e@gmx.net>.
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/misc/guc.c,v 1.166 2003/11/07 21:27:38 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/misc/guc.c,v 1.167 2003/11/13 00:40:01 wieck Exp $
  *
  *--------------------------------------------------------------------
  */
@@ -73,6 +73,7 @@ extern int	CheckPointTimeout;
 extern int	CommitDelay;
 extern int	CommitSiblings;
 extern char *preload_libraries_string;
+extern int	BufferStrategyStatInterval;
 
 #ifdef HAVE_SYSLOG
 extern char *Syslog_facility;
@@ -1190,6 +1191,15 @@ static struct config_int ConfigureNamesInt[] =
 		-1, -1, INT_MAX / 1000, NULL, NULL
 	},
 
+	{
+		{"buffer_strategy_status_interval", PGC_POSTMASTER, RESOURCES_MEM,
+			gettext_noop("Interval to report buffer strategy status in seconds"),
+			NULL
+		},
+		&BufferStrategyStatInterval,
+		0, 0, 600, NULL, NULL
+	},
+
 	/* End-of-list marker */
 	{
 		{NULL, 0, 0, NULL, NULL}, NULL, 0, 0, 0, NULL, NULL

src/backend/utils/misc/postgresql.conf.sample

@@ -58,6 +58,7 @@
 #shared_buffers = 1000		# min 16, at least max_connections*2, 8KB each
 #sort_mem = 1024		# min 64, size in KB
 #vacuum_mem = 8192		# min 1024, size in KB
+#buffer_strategy_status_interval = 0	# 0-600 seconds
 
 # - Free Space Map -
 

src/include/miscadmin.h

@@ -12,7 +12,7 @@
  * Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: miscadmin.h,v 1.134 2003/09/24 18:54:01 tgl Exp $
+ * $Id: miscadmin.h,v 1.135 2003/11/13 00:40:01 wieck Exp $
  *
  * NOTES
  *	  some of the information in this file should be moved to
@@ -96,6 +96,13 @@ extern void ProcessInterrupts(void);
 		CritSectionCount--; \
 	} while(0)
 
+#define PG_DELAY(_msec) \
+{ \
+	struct timeval delay; \
+	delay.tv_sec = (_msec) / 1000; \
+	delay.tv_usec = ((_msec) % 1000) * 1000; \
+	(void) select(0, NULL, NULL, NULL, &delay); \
+}
 
 /*****************************************************************************
  *	  globals.h --															 *

src/include/storage/buf_internals.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: buf_internals.h,v 1.61 2003/08/04 02:40:14 momjian Exp $
+ * $Id: buf_internals.h,v 1.62 2003/11/13 00:40:02 wieck Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -72,17 +72,29 @@ typedef struct buftag
 	(a)->rnode = (xx_reln)->rd_node \
 )
 
+#define BUFFERTAG_EQUALS(a,xx_reln,xx_blockNum) \
+( \
+	(a)->rnode.tblNode == (xx_reln)->rd_node.tblNode && \
+	(a)->rnode.relNode == (xx_reln)->rd_node.relNode && \
+	(a)->blockNum == (xx_blockNum) \
+)
+#define BUFFERTAGS_EQUAL(a,b) \
+( \
+	(a)->rnode.tblNode == (b)->rnode.tblNode && \
+	(a)->rnode.relNode == (b)->rnode.relNode && \
+	(a)->blockNum == (b)->blockNum \
+)
+
 /*
  *	BufferDesc -- shared buffer cache metadata for a single
  *				  shared buffer descriptor.
  */
 typedef struct sbufdesc
 {
-	Buffer		freeNext;		/* links for freelist chain */
-	Buffer		freePrev;
+	Buffer		bufNext;		/* link in freelist chain */
 	SHMEM_OFFSET data;			/* pointer to data in buf pool */
 
-	/* tag and id must be together for table lookup (still true?) */
+	/* tag and id must be together for table lookup */
 	BufferTag	tag;			/* file/block identifier */
 	int			buf_id;			/* buffer's index number (from 0) */
 
@@ -107,6 +119,7 @@ typedef struct sbufdesc
 
 #define BufferDescriptorGetBuffer(bdesc) ((bdesc)->buf_id + 1)
 
+
 /*
  * Each backend has its own BufferLocks[] array holding flag bits
  * showing what locks it has set on each buffer.
@@ -167,14 +180,19 @@ extern long int LocalBufferFlushCount;
 /*freelist.c*/
 extern void PinBuffer(BufferDesc *buf);
 extern void UnpinBuffer(BufferDesc *buf);
-extern BufferDesc *GetFreeBuffer(void);
-extern void InitFreeList(bool init);
+extern BufferDesc *StrategyBufferLookup(BufferTag *tagPtr, bool recheck);
+extern BufferDesc *StrategyGetBuffer(void);
+extern void StrategyReplaceBuffer(BufferDesc *buf, Relation rnode, BlockNumber blockNum);
+extern void StrategyInvalidateBuffer(BufferDesc *buf);
+extern void StrategyHintVacuum(bool vacuum_active);
+extern int StrategyDirtyBufferList(int *buffer_dirty, int max_buffers);
+extern void StrategyInitialize(bool init);
 
 /* buf_table.c */
-extern void InitBufTable(void);
-extern BufferDesc *BufTableLookup(BufferTag *tagPtr);
-extern bool BufTableDelete(BufferDesc *buf);
-extern bool BufTableInsert(BufferDesc *buf);
+extern void InitBufTable(int size);
+extern int BufTableLookup(BufferTag *tagPtr);
+extern bool BufTableInsert(BufferTag *tagPtr, Buffer buf_id);
+extern bool BufTableDelete(BufferTag *tagPtr);
 
 /* bufmgr.c */
 extern BufferDesc *BufferDescriptors;